Through data collection of previous incidents, AI can provide better insights on future collisions.   -  Photo: Aayush Srivastava

Through data collection of previous incidents, AI can provide better insights on future collisions. 

Photo: Aayush Srivastava

Traffic collisions, and the ability to prevent them, have been an ongoing concern within the fleet industry. According to the U.S. Department of Transportation and the National Highway Traffic Safety Administration, the number of traffic fatalities in the U.S. was 36,096 in 2019, which is more than 11.17 deaths per 100,000 population.

Finding more ways to keep pedestrians, bicyclists, and those in vehicles safe has changed the evolution of both city infrastructure, like traffic sensors, and the way vehicles are manufactured with features such as automatic emergency braking. Now artificial intelligence is taking road safety to a new level through data collection and collision predictions. And while AI hasn't become the end-all for safety solutions, it is giving us a better insight into what we can do better for those on and off the road. 

Tracking Patterns and Collisions

Multiple AI tools are already in place that allow cities to track patterns in collisions and risk incidents. Berkeley, for example, has the DeepDrive Industrial Consortium that looks at automotive perception and new ways to keep drivers safe. Nauto, AI-powered fleet management software, works by using a dual-facing camera plus external sensors to detect drowsy and distracted driving as well as in-cabin and external risks, all in real time.

"AI is really just beginning to come into traffic safety," said Nauto CEO Stefan Heck. "There is some work out there now that looks at data to try and find spots predictively, but it's just in its infancy."

Heck notes that Nauto's goal is to have driver's self correct through the application of computer vision to show risks being taken, providing feedback, and showing this information to the fleet and the driver. 

"It turns out most of the driving risks that you take are unconscious," Heck said. " You find that the vast majority of risks are risks that the drivers are not aware of."

One example is bicycle-related accidents. A study performed in Long Beach involving bicycle accidents found that 63% of all crashes were caused by a motorist's failure to yield the right-of-way, turning or both.

AI Interactions with Drivers

AI can be a powerful tool because, according to Heck, it is essentially detecting risks that could come from a situation that a driver may not feel is a risk. He uses drivers checking their phones while at a stop light as an example. 

"For most people, it doesn't intuitively make sense that 'if I'm at a stop sign or red light, why can't I look at my phone?'" Heck said. "You're not moving but you're at risk of getting rear-ended."

But once AI is being used in a vehicle, Heck said that they can see 80% of drivers changing their behavior leading to an 80% reduction of risks in a single week. This is because of the feedback provided once the AI is deployed. 

"They already know how to be good drivers," Heck said. "They're just not aware of some of these risks that are taken."

Heck also mentioned Mobileye, a driver-assist software company, that had been focusing on the scenario of rear-ending a car. This was due to the frequency that these types of collisions happened, according to Heck. 

"The most common claims event we see is backing into things," Heck said. "So that AI is now ramping up very quickly."

Heck believes that over the next five or six years, most vehicles will have some sort of AI for awareness and collision situations, whether it is built in or an option. 

Anonymizing data that comes in is also a focus for Heck and Nauto, which brings in information to map where risks are without linking back to who those drivers are. He uses the San Francisco Bay area as an example, where the system has mapped different highway exits and looked at how risky they are. That data shows there are some exits that are shown to be very safe with no high-risk behaviors. Others were found to be extremely unsafe.

"Interestingly, when we've looked at some of those, they have had historical collisions as well so would have been notable even in the past," Heck noted. However, he adds that from previous conversations with fleet leaders, he has found there aren't always the tools or people to process traffic and collision data. 

"There's still some translation needed, I think from the richness of the data science into something really practical that a city can get adopted," Heck said.

Calculating Risks

Heck explained that what he sees in Nauto's data is the accumulation of risks. It's speeding, then speeding when there's a traffic jam ahead, then speeding, tailgating, and checking your phone the risk is three times higher. And it's because of this that he said these risks have to be looked at in a geometric, rather than linear way. 

"If I just alert you for every risk, it gets very annoying, and people get fed up," Heck said "That's why they turn off lane departure warning because half the time it's alerting them for something that is not really a risk."

However, Heck said in high-risk situations, like one someone is looking at their phone while a pedestrian is crossing, there is zero tolerance and drivers should be alerted. 

"There are certainly video telematics and other camera systems out there that just record everything or just beep at you all the time. Those are not very good solutions," Heck said. "That's not where the automakers are going. They're really looking at how serious the risk is."

Solutions for Fleet Managers

For fleet managers trying to best stay on top of all of their vehicles, the high amount of data coming in, from telematics or video for example, can be overwhelming and time consuming. That video may get sent to a third party for review, which can be a security concern. 

With AI, drivers can get real-time feedback eliminating the need to wait for someone to watch a video. Warnings will be given for pedestrians, speeding, checking phones and more. And this allows for self correction. It's this that Heck said they have seen that 80% of drivers in their fleets automatically become better on their own.

"There's no the fleet manager, the supervisor doesn't have to do anything, and they can then focus on the bottom four or five percent of drivers," Heck said. "We do see there's such a small number of drivers in every fleet that do need more training or need ride alongs or coaching, and so you can then focus the supervisor's effort on them."

Check out these examples of how people, technology, and equipment are boosting safety
0 Comments